PLS
G7
2024-05-05
VARIABLE SELECTION
df <- read_excel("df_definitivo-2.xlsx", sheet = "Copia de datos")
rownames(df) = df$id## Warning: Setting row names on a tibble is deprecated.df$Afectacion_ganglionar = as.numeric(df$Afectacion_ganglionar)
df$Afectacion_metastasica = as.numeric(df$Afectacion_metastasica)
df$Tamaño_tumor = as.numeric(df$Tamaño_tumor)
df$Grado_tox = as.numeric(df$Grado_tox)
df$SII_pre = log(df$SII_pre)
df$SII_1C = log(df$SII_1C)
df$SII_2C = log(df$SII_2C)
df$SII_1eval = log(df$SII_1eval)
variables_inutiles = c("Anciano", "Peso", "Talla", "SG", "SG_cens", "SLP", "SLP_cens", "Tipo_tox", "NLR2C_corte4o5")
df2 = select(df, -all_of(variables_inutiles))PLS
SPECIFIC VARIABLES FIRST EVAL
#colnames(df2)
X1.1 = df2[,2:49]
X1.2 = df2[,61:67]
X = bind_cols(X1.1, X1.2)
Y = as.matrix(df2['pri_eval_num_ok'])SPECIFIC VARIABLES BEST RESPONSE
#colnames(df2)
X2 = df2[,2:67]
Y2 = as.matrix(df2["mejor_resp_num_ok"])SCALING
mypls = opls(x = X, y = Y, predI = NA, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS
## 34 samples x 55 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 0.147 0.526 0.295 0.489 1 0 0.15 0.05NUMBER OF COMPONENTS SELECTION
mypls_2 = opls(x = X, y = Y, predI = 33, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS
## 34 samples x 55 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 1 1 1 Inf 33 0 1.05 1.05plot(1:33, mypls_2@modelDF$`R2Y(cum)`, type = "o", pch = 16, col = "blue3",
lwd = 2, xlab = "Components", ylab = "",
main = "PLS model for predicting First Evaluation", ylim = c(-1,1))
lines(1:33, mypls_2@modelDF$`Q2(cum)`, type = "o", pch = 16, col = "red3",
lwd = 2)
abline(h = 0, col = "red3", lty = 2)
legend("bottomleft", c("R2Y", "Q2"), lwd = 2,
col = c("blue3", "red3"), bty = "n")
mypls = opls(x = X, y = Y, predI = 2, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS
## 34 samples x 55 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 0.242 0.7 0.178 0.395 2 0 0.1 0.05EXTREME INDIVIDUALS STUDY
Para estudiar los datos anómalos haremos uso de la T2 de Hotelling y la Suma de Cuadrados Residual.
En el PCA observabamos que el individuo 11 es un dato atipico extremo. Pero igualmente lo representaremos en el gráfico de la T2 de hotelling.
misScores = mypls@scoreMN
varT = apply(misScores, 2, var)
miT2 = colSums(t(misScores**2) / varT)
N = nrow(X)
A = 2
F95 = A*(N**2 - 1)/(N*(N - A)) * qf(0.95, A, N-A); F95## [1] 6.994835F99 = A*(N**2 - 1)/(N*(N - A)) * qf(0.99, A, N-A); F99## [1] 11.32992plot(1:length(miT2), miT2, type = "l", xlab = "pacientes", ylab = "T2",
main = "PLS: T2-Hotelling", ylim = c(0,15))
abline(h = F95, col = "orange", lty = 2, lwd = 2)
abline(h = F99, col = "red3", lty = 2, lwd = 2)
myT = mypls@scoreMN
myP = mypls@loadingMN
myE = scale(X) - myT%*%t(myP)
mySCR = rowSums(myE^2) # SPE
plot(1:length(mySCR), mySCR, type = "l", main = "PLS: SCR Distance to the model",
ylab = "SCR", xlab = "pacientes", ylim = c(0,100))
g = var(mySCR)/(2*mean(mySCR))
h = (2*mean(mySCR)^2)/var(mySCR)
chi2lim = g*qchisq(0.95, df = h)
abline(h = chi2lim, col = "orange", lty = 2)
chi2lim99 = g*qchisq(0.99, df = h)
abline(h = chi2lim99, col = "red3", lty = 2)
GRPAHICS
plot(x = mypls, typeVc = "x-score", parCompVi = c(1, 2), parLabVc = rownames(X), parPaletteVc = NA, parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls, typeVc = "x-loading",
parCexN = 0.8, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls, typeVc = "xy-weight",
parCexN = 0.5, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
VIP = data.frame(sort(mypls@vipVn))
bottom_10 <- data.frame(variables = rownames(VIP)[1:10], vip = VIP[1:10, ])
top_10 <- data.frame(variables = rownames(VIP)[46:55], vip = VIP[46:55, ])
grafico_barras <- ggplot(top_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Top 10 VIP variables")
grafico_barras2 <- ggplot(bottom_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Bottom 10 VIP variables")
print(grafico_barras)
print(grafico_barras2)
EVALUATION METRICS FOR REGRESSION
Ypred = predict(mypls)
residuos = Y-Ypred
myRMSE = sqrt(colMeans(residuos^2))
CVrmse = myRMSE/colMeans(Y)
myRMSE ## pri_eval_num_ok
## 0.3771181CVrmse## pri_eval_num_ok
## 0.2137002for (i in 1:ncol(Y)) {
plot(Y[,i], Ypred[,i], asp = 1, main = colnames(Y)[i],
xlab = "observado", ylab = "predicho")
abline(a=0, b=1, col = "red3", lwd = 2, lty = 2)
}
SCALING
mypls2 = opls(x = X2, y = Y2, predI = NA, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS
## 34 samples x 66 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 0.172 0.595 0.416 0.593 1 0 0.05 0.05mypls_2.2 = opls(x = X2, y = Y2, predI = 33, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS
## 34 samples x 66 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 1 1 1 Inf 33 0 1.05 1.05NUMBER OF COMPONENTS SELECTION
plot(1:33, mypls_2.2@modelDF$`R2Y(cum)`, type = "o", pch = 16, col = "blue3",
lwd = 2, xlab = "Components", ylab = "",
main = "PLS model for predicting Best Response", ylim = c(-1,1))
lines(1:33, mypls_2.2@modelDF$`Q2(cum)`, type = "o", pch = 16, col = "red3",
lwd = 2)
abline(h = 0, col = "red3", lty = 2)
legend("bottomleft", c("R2Y", "Q2"), lwd = 2,
col = c("blue3", "red3"), bty = "n")
mypls_2 = opls(x = X2, y = Y2, predI = 4, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS
## 34 samples x 66 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 0.475 0.915 0.381 0.286 4 0 0.05 0.05EXTREME INDIVIDUALS STUDY
Para estudiar los datos anómalos haremos uso de la T2 de Hotelling y la Suma de Cuadrados Residual.
En el PCA observabamos que el individuo 11 es un dato atipico extremo. Pero igualmente lo representaremos en el gráfico de la T2 de hotelling.
misScores = mypls_2@scoreMN
varT = apply(misScores, 2, var)
miT2 = colSums(t(misScores**2) / varT)
N = nrow(X2)
A = 2
F95 = A*(N**2 - 1)/(N*(N - A)) * qf(0.95, A, N-A); F95## [1] 6.994835F99 = A*(N**2 - 1)/(N*(N - A)) * qf(0.99, A, N-A); F99## [1] 11.32992plot(1:length(miT2), miT2, type = "l", xlab = "pacientes", ylab = "T2",
main = "PLS: T2-Hotelling", ylim = c(0,15))
abline(h = F95, col = "orange", lty = 2, lwd = 2)
abline(h = F99, col = "red3", lty = 2, lwd = 2)
El individuo 11 sigue saliendo como dato extremo, aun realizando
transformaciones logarÃtmicas.
myT = mypls_2@scoreMN
myP = mypls_2@loadingMN
myE = scale(X2) - myT%*%t(myP)
mySCR = rowSums(myE^2) # SPE
plot(1:length(mySCR), mySCR, type = "l", main = "PLS: SCR Distance to the model",
ylab = "SCR", xlab = "pacientes", ylim = c(0,100))
g = var(mySCR)/(2*mean(mySCR))
h = (2*mean(mySCR)^2)/var(mySCR)
chi2lim = g*qchisq(0.95, df = h)
abline(h = chi2lim, col = "orange", lty = 2)
chi2lim99 = g*qchisq(0.99, df = h)
abline(h = chi2lim99, col = "red3", lty = 2)
GRAPHICS
plot(x = mypls_2, typeVc = "x-score", parCompVi = c(1, 2), parLabVc = rownames(X), parPaletteVc = NA, parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls_2, typeVc = "x-score", parCompVi = c(3, 4), parLabVc = rownames(X), parPaletteVc = NA, parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls_2, typeVc = "x-score", parCompVi = c(1,3), parLabVc = rownames(X), parPaletteVc = NA, parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls_2, typeVc = "x-score", parCompVi = c(1, 4), parLabVc = rownames(X), parPaletteVc = NA, parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls_2, typeVc = "x-loading",
parCexN = 0.8, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls_2, typeVc = "xy-weight",
parCexN = 0.5, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
plot(x = mypls_2, typeVc = "xy-weight",
parCexN = 0.5, parCompVi = c(3, 4), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
VIP2 = data.frame(sort(mypls_2@vipVn))
bottom_10 <- data.frame(variables = rownames(VIP2)[1:10], vip = VIP[1:10, ])
top_10 <- data.frame(variables = rownames(VIP2)[46:55], vip = VIP[46:55, ])
grafico_barras <- ggplot(top_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Top 10 VIP variables")
grafico_barras2 <- ggplot(bottom_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Bottom 10 VIP variables")
print(grafico_barras)
print(grafico_barras2)
# EVALUATION METRICS FOR REGRESSION
Ypred2 = predict(mypls_2)
residuos2 = Y2-Ypred2
myRMSE2 = sqrt(colMeans(residuos2^2))
CVrmse2 = myRMSE2/colMeans(Y2)
myRMSE2 ## mejor_resp_num_ok
## 0.2638897CVrmse2## mejor_resp_num_ok
## 0.1950489El RMSE expresa la diferencia entre los valores predichos por nuestro modelo y los observados. Es decir, en promedio el valor difiere +/- 0.2638 del valor real. Es normal porque los valores que se observan son números enteros (1, 2 y 3) y los valores que se predicen son números reales entre el 1 y el 3.
El CVrmse que obtenemos es de 0.195 lo que significa que las predicciones del modelo tienen un error del 19.5% en relación con la escala de los datos originales. Con este valor podriamos decir que el modelo tiene un rendimiento moderado-aceptable.
for (i in 1:ncol(Y)) {
plot(Y2[,i], Ypred2[,i], asp = 1, main = colnames(Y2)[i],
xlab = "observado", ylab = "predicho")
abline(a=0, b=1, col = "red3", lwd = 2, lty = 2)
}
PLS-DA
SPECIFIC VARIABLES FIRST EVAL
X1.1 = df2[,2:49]
X1.2 = df2[,61:67]
X = bind_cols(X1.1, X1.2)
Y = df2$pri_eval_num_ok
Y = as.matrix(as.factor(Y))SPECIFIC VARIABLES BEST RESPONSE
X2 = df2[,2:67]
Y2 = df2$mejor_resp_num_ok
Y2 = as.matrix(as.factor(Y2))SCALING
myplsda = opls(x = X, y = Y, predI = 33, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS-DA
## 34 samples x 55 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 1 1 1 Inf 33 0 1.05 1.05NUMBER OF COMPONENTS SELECTION
plot(1:33, myplsda@modelDF$`R2Y(cum)`, type = "o", pch = 16, col = "blue3",
lwd = 2, xlab = "Components", ylab = "", ylim = c(-0.5, 1),
main = "PLS-DA model for predicting First Evaluation")
lines(1:33, myplsda@modelDF$`Q2(cum)`, type = "o", pch = 16, col = "red3",
lwd = 2)
abline(h = 0, col = "red3", lty = 2)
legend("bottomleft", c("R2Y", "Q2"), lwd = 2,
col = c("blue3", "red3"), bty = "n")
myplsda = opls(x = X, y = Y, predI = 2, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS-DA
## 34 samples x 55 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 0.25 0.442 -0.14 0.361 2 0 0.4 0.25EXTREME INDIVIDUALS STUDY
misScores = myplsda@scoreMN
varT = apply(misScores, 2, var)
miT2 = colSums(t(misScores**2) / varT)
N = nrow(X)
A = 3
F95 = A*(N**2 - 1)/(N*(N - A)) * qf(0.95, A, N-A);
F99 = A*(N**2 - 1)/(N*(N - A)) * qf(0.99, A, N-A);
plot(1:length(miT2), miT2, type = "l", xlab = "Patients", ylab = "T2",
main = "PLS: T2-Hotelling")
abline(h = F95, col = "orange", lty = 2, lwd = 2)
abline(h = F99, col = "red3", lty = 2, lwd = 2)
myT = myplsda@scoreMN
myP = myplsda@loadingMN
myE = scale(X) - myT%*%t(myP)
mySCR = rowSums(myE^2) # SPE
plot(1:length(mySCR), mySCR, type = "l", main = "SCR",
xlab = "Patients")
g = var(mySCR)/(2*mean(mySCR))
h = (2*mean(mySCR)^2)/var(mySCR)
chi2lim = g*qchisq(0.95, df = h)
abline(h = chi2lim, col = "orange", lty = 2)
chi2lim99 = g*qchisq(0.99, df = h)
abline(h = chi2lim99, col = "red3", lty = 2)
GRAPHICS
plot(x = myplsda, typeVc = "x-score",
parCexN = 0.5, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
plot(x = myplsda, typeVc = "xy-weight",
parCexN = 0.5, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
VIP = data.frame(sort(myplsda@vipVn))
bottom_10 <- data.frame(variables = rownames(VIP)[1:10], vip = VIP[1:10, ])
top_10 <- data.frame(variables = rownames(VIP)[46:55], vip = VIP[46:55, ])
grafico_barras <- ggplot(top_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Top 10 VIP variables")
grafico_barras2 <- ggplot(bottom_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Bottom 10 VIP variables")
print(grafico_barras)
print(grafico_barras2)
EVALUATION METRICS FOR CLASSIFICATION
myplsda_pred = predict(myplsda)
actual = factor(df2$pri_eval_num_ok)
myplsda_pred <- factor(myplsda_pred, levels = levels(actual))
confusionMatrix(myplsda_pred, actual, positive = "3")## Confusion Matrix and Statistics
##
## Reference
## Prediction 1 2 3
## 1 11 4 0
## 2 2 11 0
## 3 0 1 5
##
## Overall Statistics
##
## Accuracy : 0.7941
## 95% CI : (0.621, 0.913)
## No Information Rate : 0.4706
## P-Value [Acc > NIR] : 0.0001154
##
## Kappa : 0.6708
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: 1 Class: 2 Class: 3
## Sensitivity 0.8462 0.6875 1.0000
## Specificity 0.8095 0.8889 0.9655
## Pos Pred Value 0.7333 0.8462 0.8333
## Neg Pred Value 0.8947 0.7619 1.0000
## Prevalence 0.3824 0.4706 0.1471
## Detection Rate 0.3235 0.3235 0.1471
## Detection Prevalence 0.4412 0.3824 0.1765
## Balanced Accuracy 0.8278 0.7882 0.9828Elegimos la clase 3 como la positiva porque al creemos que es más importante detectar los pacientes que están teniendo una progresión de la enfermedad que los que la enfermedad está progresando o al menos estable. Podemos ver que el modelo si que encuentra aquellos que están empeorando (clase 3).
El Ãndice de kappa nos indica que las predicciones están cerca de haberse realizado por casualidad. Podriamos decir que esto se debe a la clase 2 sobre todo donde se encuentran pacientes con una ligera pseudoprogresión y la enfermedad estable, es decir, que están entre la clase 1 y 3 tres totalmente.
Sin embargo, si miramos el valor de balanced accuracy nos sugiere que el modelo no está haciendo tan malas predicciones y que al menos si nos centramos en predecir los que son de la clase 3 lo está haciendo correctamente.
De este modelo podemos extraer como conclusión que quizas habrÃa que redefinir las clases propuestas ya que podrÃan haber más o menos clases que nos ayudarÃan a separar mejor los pacientes con este PLS.
SCALING
myplsda2 = opls(x = X2, y = Y2, predI = NA, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS-DA
## 34 samples x 66 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 0.188 0.203 0.0777 0.376 1 0 0.05 0.05NUMBER OF COMPONENTS SELECTION
myplsda2 = opls(x = X2, y = Y2, predI = 33, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS-DA
## 34 samples x 66 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 1 1 1 Inf 33 0 1.05 1.05plot(1:33, myplsda2@modelDF$`R2Y(cum)`, type = "o", pch = 16, col = "blue3", lwd = 2, xlab = "Components", ylab = "", main = "PLS-DA model for predicting Best Response", , ylim = c(-0.5, 1))
lines(1:33, myplsda2@modelDF$`Q2(cum)`, type = "o", pch = 16, col = "red3", lwd = 2)
abline(h = 0, col = "red3", lty = 2)
legend("bottomleft", c("R2Y", "Q2"), lwd = 2,
col = c("blue3", "red3"), bty = "n")
myplsda2 = opls(x = X2, y = Y2, predI = 2, crossvalI = nrow(X), scaleC = "standard", fig.pdfC = "none")## PLS-DA
## 34 samples x 66 variables and 1 response
## standard scaling of predictors and response(s)
## R2X(cum) R2Y(cum) Q2(cum) RMSEE pre ort pR2Y pQ2
## Total 0.327 0.354 0.115 0.347 2 0 0.05 0.05EXTREME INDIVIDUALS STUDY
misScores = myplsda2@scoreMN
varT = apply(misScores, 2, var)
miT2 = colSums(t(misScores**2) / varT)
N = nrow(X2)
A = 4
F95 = A*(N**2 - 1)/(N*(N - A)) * qf(0.95, A, N-A);
F99 = A*(N**2 - 1)/(N*(N - A)) * qf(0.99, A, N-A);
plot(1:length(miT2), miT2, type = "l", xlab = "Patients", ylab = "T2",
main = "PLS: T2-Hotelling",ylim=c(0,20))
abline(h = F95, col = "orange", lty = 2, lwd = 2)
abline(h = F99, col = "red3", lty = 2, lwd = 2)
myT = myplsda2@scoreMN
myP = myplsda2@loadingMN
myE = scale(X2) - myT%*%t(myP)
mySCR = rowSums(myE^2) # SPE
plot(1:length(mySCR), mySCR, type = "l", main = "SCR",
xlab = "Patients")
g = var(mySCR)/(2*mean(mySCR))
h = (2*mean(mySCR)^2)/var(mySCR)
chi2lim = g*qchisq(0.95, df = h)
abline(h = chi2lim, col = "orange", lty = 2)
chi2lim99 = g*qchisq(0.99, df = h)
abline(h = chi2lim99, col = "red3", lty = 2)
GRAPHICS
plot(x = myplsda2, typeVc = "x-score",
parCexN = 0.8, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
plot(x = myplsda2, typeVc = "xy-weight",
parCexN = 0.7, parCompVi = c(1, 2), parPaletteVc = NA,
parTitleL = TRUE, parCexMetricN = NA)
VIP = data.frame(sort(myplsda2@vipVn))
bottom_10 <- data.frame(variables = rownames(VIP)[1:10], vip = VIP[1:10, ])
top_10 <- data.frame(variables = rownames(VIP)[46:55], vip = VIP[46:55, ])
grafico_barras <- ggplot(top_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Top 10 VIP variables")
grafico_barras2 <- ggplot(bottom_10, aes(x = reorder(variables, vip), y = vip)) + geom_bar(stat = "identity", fill = "skyblue") + theme(axis.text.x = element_text(angle = 45, hjust = 1)) + labs(x = "Variable", y = "VIP", title = "Bottom 10 VIP variables")
print(grafico_barras)
print(grafico_barras2)
EVALUATION METRICS FOR CLASSIFICATION
mypred2 = predict(myplsda2)
actual = factor(df2$mejor_resp_num_ok)
mypred2 <- factor(mypred2, levels = levels(actual))
confusionMatrix(mypred2, actual, positive = "3")## Confusion Matrix and Statistics
##
## Reference
## Prediction 0 1 2 3
## 0 0 0 0 0
## 1 4 17 3 0
## 2 0 0 3 0
## 3 1 0 1 5
##
## Overall Statistics
##
## Accuracy : 0.7353
## 95% CI : (0.5564, 0.8712)
## No Information Rate : 0.5
## P-Value [Acc > NIR] : 0.004521
##
## Kappa : 0.5578
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: 0 Class: 1 Class: 2 Class: 3
## Sensitivity 0.0000 1.0000 0.42857 1.0000
## Specificity 1.0000 0.5882 1.00000 0.9310
## Pos Pred Value NaN 0.7083 1.00000 0.7143
## Neg Pred Value 0.8529 1.0000 0.87097 1.0000
## Prevalence 0.1471 0.5000 0.20588 0.1471
## Detection Rate 0.0000 0.5000 0.08824 0.1471
## Detection Prevalence 0.0000 0.7059 0.08824 0.2059
## Balanced Accuracy 0.5000 0.7941 0.71429 0.9655